Live demo · a company that does not exist

Ask a company a question its own systems cannot answer.

Nortia Labs is invented — 92 people, 123 records, 329 relationships across ten departments. The renewal lives in the CRM, the incident in the helpdesk, the bug in the tracker, the contract in a drive and the invoice in the ledger. Nothing here can answer why did that renewal slip. The graph can.

nortia-labs/graphify-out/graph.json123 nodes · 329 edges · 10 communities

drag a record · pinch to zoom · click a department to isolate it

$ brain ask Ask this company anything. The graph holds accounts, deals, incidents, contracts, decisions and people.

Try one of these

The plate is live. Colour groups the ten departments; the ring around a record means the question's own words found it, and everything else was reached by following the edges.

I.

Questions worth asking

Every one of these is ordinary, and every one of them takes five systems to answer by hand.

Across departments

One answer that no single system holds.

Risk and retention

The things that only look small until you follow the edges.

People and process

The knowledge that walks out of the building.

II.

What happens when you ask

  1. 01

    Extract

    Every source system is dumped to plain Markdown, then read once into a graph: records become nodes, the relationships between them become edges, and each edge is graded by how directly it was stated. Ten communities fall out of the structure — they are the departments.

  2. 02

    Retrieve

    Your question scores every record, and the best few become seeds. Then the search walks outwards along the edges for two hops, following the explanatory ones hardest — blocked by, caused by, depends on. That walk is what reaches the bug from a question about a renewal.

  3. 03

    Answer

    Only those records go to the model, with the edges between them and an instruction to cite the id behind every claim. No embeddings, no vector store — the structure was already there.

retrieval
$ graphify query "why did the Vantor renewal slip?"
seeds 6 records matched the question's own words
walk 2 hops along blocks, caused_by, depends_on
22 records across 7 departments · 3ms
 
sent to the model as context, with the edges between them

The same walk the plate animates, as the CLI prints it. graphify ships the query; brain is the wrapper that builds the graph in the first place.

{
  "id": "n028",
  "label": "INC-2291 — Vantor nightly sync dropping records",
  "norm_label": "inc 2291 vantor nightly sync dropping records",
  "file_type": "document",
  "source_file": "raw/zendesk/events/inc-2291-....md",
  "community": 1,
  "community_name": "Customer Success"
}

One record, exactly as it sits in the file. This is a NetworkX node_link_data graph — the artifact graphify extract writes for a real company, not a shape invented for this page.

Your company, not this one

Nortia Labs took an afternoon to invent. Yours already exists — it is just spread across nine systems.